We propose an individual based model to generate commuting networks from data giving numbers of in and out commuters of every locations of a region. It is inspired from the gravity model. We exhibit from the calibration of 80 various case studies that the only remaining parameter can be suppressed since it follows a law depending on the average surface of the locations. We compare our model to the only one other universal existing approach (Simini et al, 2012) at various scales. We show our model performs better for small locations (in terms of surface and/or number of inhabitants).